Which workers and jobs will be most affected by machine learning?

Machine learning, a form of artificial intelligence, is increasingly being used by Canadian firms to drive innovation and raise productivity. Because machine learning can adapt and generate outputs with increasing independence, this technology can be used to perform physical or cognitive job tasks across a broad range of industries and occupations.

Previous technological transformations—for example, the automation of work—have tended to negatively affect workers from socially and economically disadvantaged groups the most. This study set out to examine which worker groups in Canada are more likely to be affected by machine learning—that is, they’re in jobs in which a high proportion of tasks could be performed by this type of technology.

To estimate “machine learning exposure” (the degree to which job tasks could be performed by machine learning) among Canadian jobs, the researchers used an existing U.S.-based measure that scored 1,000 U.S. occupations. The team applied these scores onto equivalent occupations in the Canadian labour market. Then, using data from eight years of Statistics Canada’s Labour Force Survey (LFS)—from 2013 to 2019, and 2022—the team divided the sample of Canadian workers into three groups of roughly equal sizes: those in occupations with high, medium and low levels of machine learning exposure.

The team also used LFS data to compare the three groups along sociodemographic factors. The team focused on three factors in particular—education, hourly wages, and job skill requirements –and analyzed how these related to a worker’s likelihood of being in jobs with high or low machine learning exposure. Finally, the team examined how these relationships differed for men and women.

The study found that every Canadian occupation included at least some job tasks that could be done by machine learning—but no single job could be done completely by machine learning.

Machine learning exposure varied with worker education, worker earnings and the skill requirements of the job. But these differences didn’t follow the same ordered pattern as was often seen in past technological changes. With machine learning, it was not always the workers at the low end of the socioeconomic spectrum who were most affected.

In terms of education, workers with a college or bachelor's degree or more tended to be in occupations with high machine learning exposure. Workers with some post-secondary or less (the group with the lowest educational attainment) tended to be in jobs with low machine learning exposure.

In terms of earnings, the worker groups with high exposure to machine learning were neither the top nor the bottom earners, but the group in the middle. And in terms of job skills requirements, the two groups most exposed to machine learning were the group with the second-lowest skills (semi-skilled occupations) and the group with the highest skills (managerial occupations). Workers in the middle groups tended to have the lowest exposure to machine learning.

The study also found important differences between men and women. More women workers (45 per cent) than men (28 per cent) were in occupations with high machine learning exposure. Additional differences—some unexpected—were found between women and men when it came to the relationships between the three studied factors and machine learning exposure.

For example, while men with a college or university degree or with higher wages were more likely to be in jobs with high machine learning exposure than men less education or lower wages, that pattern was not the same for women. Women with higher education and higher wages had lower exposure to machine learning than women with less education and less pay.

Although all workers should be prepared for machine learning in their occupations to some degree, policy and programs to address any impacts should focus on the worker groups that are most likely to be affected by the technology.

It is not clear based on this study whether workers in jobs with high exposure to machine learning will experience advantages or disadvantages from this technology. However, the disproportionate impacts of machine learning across different worker groups raise concerns about potential inequities that could emerge with the rise of machine learning and other forms of artificial intelligence adoption in the labour market.

This study is one of the few that provides an overview of potential machine learning exposure in different occupations across Canada, and how this exposure varies according to sociodemographic and occupational characteristics.

The results of this study represent estimates and should not be interpreted as the amount of actual use or adoption of machine learning within these occupations in the coming years. The researchers were unable to speculate on the specific impacts machine learning will have on workers and job performance. Additionally, the suitability for machine learning measurement may not encompass all forms of artificial intelligence and work automation that may continue to be developed as the technology rapidly advances.